to characterize a LI to be selected as a head in a way that distinguishes it from,
for example, a phrase? This is not obvious in the theory of bare phrase structure
where the notion of projection is eliminated. We would like to suggest that
formal features, as computational atoms, play a vital role in determining a label
(and hence a head) for SOs.
Consider fi rst non-agreeing or dependent-marking languages, such as Japanese.
As we presented in the above sections, external Merge forms a covariance rela-
tion, and case valuation takes place based on this relation. For example, Search 0
applies to the workspace and picks out hon ‘books’ and yomu ‘read,’ and Merge 0
forms a set of the two.
(4) M 0 ◦ S 0 (WS) = {{hon}, yomu}
According to (1a), the case feature of hon is valued as accusative. Now, let
us take the accusative case value on the nominal as a marker of an “argument,”
following the basic insight of Keenan (1999) and Keenan and Stabler (2003).
This means that the other element of the SO, yomu, turns out to be a predicate,
namely the head of the SO. In order to determine the label of the SO {{hon},
yomu}, Search 0 searches yomu which does not have a case feature, and picks
out the SO and yomu from the domain. Merge 0 forms a set {{hon, yomu} yomu}.
Then, yomu is interpreted as a head of the SO in the CI interface. In this way,
a valued case feature helps Search 0 to correctly seek a head from the domain it
applies to.^6
On the other hand, there are languages in which case assignment/valuation
does not seem to apply to nominal expressions. These are so-called head-marking
languages (Baker 1996; Nichols 1986). The prominent feature of these languages
is the fact that their grammatical relations are marked only by infl ectional affi xes,
as the following example from Mohawk shows.
(5) Sak ra-nuhwe’-s ako-[a]tya’tawi.
Sak MsS-like-HAB FSP-dress
‘Sak likes her dress’
(Baker 1996:10)
How can labels be given to SOs in these languages, given that case features
of nominal expressions play no role? It is reasonable to claim that the valued
φ-features of verbal elements such as v and C/T take the place of valued case
features in directly determining the label. That is, such features identify the verbal
expression as heads. Suppose that a derivation has reached the stage where a
subject has merged with TP. An agreement relation between the φ-features of
the subject and the φ-features of T is established via M 0 ◦ S 0 , yielding valued
φ-features of T. Search 0 applies to the SO {subject, TP} and searches the T,
bearing valued φ-features. Merge 0 forms a set of the SO and T, from which the
interpretation that T is a head of the SO is derived. In this way, the valued
φ-features of T play a role in providing labels to SOs.
52 Mihoko Zushi